Energy-efficiency improvements in computing have slowed due to the demise of Moore’s Law and metal-oxide-semiconductor field-effect-transistor (MOSFET) scaling, also known as Dennard scaling.
Along with the von Neumann bottleneck, a throughput limitation caused by computer architecture, this could affect the data-rich future that will be dominated by machine learning (ML).
In-memory computing using memristor crossbars has emerged as potential alternative, as it is likely to be more energy-efficient and doesn’t suffer from the von Neumann bottleneck. However, existing in-memory memristor-dependent methods, such as neuromorphic computing, can be affected by resistance drift leading to inaccurate output and high-energy utilization; and they’re not robust against radiation damage.
Dr. Sunny Raj, a computer science and engineering (CSE) assistant professor at Oakland University, was awarded an National Science Foundation Computer and Information Science and Engineering Research Initiation Initiative (NSF CRII) grant for IMMENSE: In-memory machine learning using sneak-paths in crossbars for robustness and energy efficiency.
This marks the third CRII grant received by Oakland University School of Engineering and Computer Science this academic year.
The project seeks to pursue a new approach to crossbar computing that transcends current neuromorphic approaches and counterintuitively leverages sneak paths in 3D crossbars of emerging devices for performing computations. The impact of energy-efficient and radiation-hardened machine-learning devices enabled by this project could be enormous because the low energy requirements of flow-based memristor crossbar computing could enable the goal of energy-efficient ML systems.
Since these devices are robust against radiation damage, they will allow their use in radiation-rich environments such as space. The project seeks to train undergraduate and graduate students in the science of flow-based crossbar computing and prepare an inclusive next-generation workforce in the area of electronic design automation. The results of the project will be publicly disseminated at conferences and workshops to ensure a wide reach to stakeholders in academia, government, and industry.
IMMENSE aims to address how to design in-memory crossbar circuits for machine-learning algorithms such as support vector machines and deep neural networks for energy-efficient predictions while being robust against resistance drift and radiation degradation.
The goal of this project is to create algorithms for mapping programs to 3D crossbars and their theoretical characterizations that explain the computing capacity of crossbars in the context of data structures, such as different types of decision diagrams used in formal methods.
The three primary objectives are:
• identify bipartite data structures as transformation targets for non-kernel ML algorithms such as random forest, linear and logistic regression
• determine novel spatial abstractions of memristor crossbar to allow efficient memristor utilization for lower energy and space utilization
• identify function composition operators for crossbar abstractions to enable mapping kernelized ML algorithms onto crossbars.
